CN104956397B - Automatic spatial context based multi-object segmentation in 3D images - Google Patents
Automatic spatial context based multi-object segmentation in 3D images Download PDFInfo
- Publication number
- CN104956397B CN104956397B CN201380072420.4A CN201380072420A CN104956397B CN 104956397 B CN104956397 B CN 104956397B CN 201380072420 A CN201380072420 A CN 201380072420A CN 104956397 B CN104956397 B CN 104956397B
- Authority
- CN
- China
- Prior art keywords
- model
- test
- data handling
- magnetic resonance
- volume
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/231—Hierarchical techniques, i.e. dividing or merging pattern sets so as to obtain a dendrogram
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/149—Segmentation; Edge detection involving deformable models, e.g. active contour models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
- G06V10/7625—Hierarchical techniques, i.e. dividing or merging patterns to obtain a tree-like representation; Dendograms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30008—Bone
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/03—Recognition of patterns in medical or anatomical images
- G06V2201/033—Recognition of patterns in medical or anatomical images of skeletal patterns
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Human Computer Interaction (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention relates to automatic spatial context based multi-object segmentation in 3D images, in particular methods and systems for automatic classification of images of internal structures of human and animal bodies. A method includes receiving (405) a magnetic resonance (MR) image testing model and determining a testing volume of the testing model that includes areas of the testing model to be classified as bone or cartilage. The method includes modifying the testing model so that the testing volume corresponds to a mean shape and a shape variation space of an active shape model and producing an initial classification of the testing volume by fitting the testing volume to the mean shape and the shape variation space.; The method includes producing (425) a refined classification of the testing volume into bone areas and cartilage areas by refining the boundaries of the testing volume with respect to the active shape model and segmenting the MR image testing model into different areas corresponding to bone areas and cartilage areas.
Description
The present invention relates to according to independent claims for the 3d images based on automatic spatial context in many objects
The method of segmentation and computer-readable medium.
Technical field
The disclosure is roughly imaged and image analysis system for area of computer aided, and similar system, including but do not limit
In nuclear magnetic resonance（MRI）, NMR (Nuclear Magnetic Resonance)-imaging（NMRI）With magnetic resonance tomography method（MRT）System（Jointly and non-row
His property, " imaging system "）.
Background technology
Imaging system can be produced, be stored, manipulating and analysis of the image, including the figure of the internal structure of human and animal's body
Picture.Improved system is to close desired.
The content of the invention
Each disclosed embodiment includes the automatic classification of the image of the internal structure for human and animal's body
Method and system.Method includes receiving magnetic resonance（MR）Image measurement model, and determine test model including test model
Skeleton to be categorized as or cartilage region test volume.The method includes changing test model so that test volume correspondence
In mean shape and the change of shape space of active shape model, and it is suitable for mean shape and shape by making test volume
Change space and produce the preliminary classification of test volume.The method is included by refining test volume with regard to active shape model
Border and produce test volume to bony areas and the classification of cartilaginous areas.The method can include storing classification
It is the categorical data being associated with MR image measurement models.The method is included MR image measurement model segments according to classification
Into the zones of different corresponding to bony areas and cartilaginous areas.
For including receiving MR image measurement moulds to the other method classified by the skeleton and cartilage in magnetic resonance image (MRI)
Type, the test model include non-classified skeleton and the cartilage portion represented by multiple voxels.The method is included by data
Processing system performs the first categorizing process using first pass random forest grader, to produce each voxel of test model
The first pass probability graph of be categorized as in a strand cartilage, shin cartilage, kneecap cartilage or background.The method is included by data processing
System performs the second categorizing process using second time random forest grader, to produce each voxel of first pass probability graph
Second time probability graph of be categorized as in a strand cartilage, shin cartilage, kneecap cartilage or background.The method is included in data processing system
Store in system corresponding to second time probability graph, the categorical data being such as associated with MR image measurement models.The method includes showing
MR image measurement models, including point sector of breakdown corresponding to each voxel for indicating MR image measurement models.
The non-feature for widely outlining the disclosure and technological merit above so that those skilled in the art can be more
Understand well ensuing detailed description.Description is formed into supplementary features of the disclosure of claimed subject matter and excellent afterwards
Point.It will be appreciated by persons skilled in the art that they can easily use disclosed concept and specific embodiment as being used for
The basis of the other structures of the identical purpose of the disclosure is carried out in modification or design.It will also be appreciated by the skilled artisan that so
Equivalent constructions without departing from the disclosure its most wide form spirit and scope.
Before following description is carried out, certain used through patent document can be advantageously illustrated
The definition of a little words or phrase：Term " including " and "comprising" and its derivative mean to include without limiting；Term "or"
Comprising, it is intended that and/or；Phrase " with ... be associated " and " associated with it " and its subform can mean bag
Include, be included in ... it is interior, and ... interconnection, comprising, be included in ... it is interior, be connected to or with ... be connected, be coupled to or
With ... coupling, with ... can communicate, with ... cooperate, interweave, juxtaposition, be close to, be attached to or with ... combination, with, have
Property having ... etc.；And term " controller " means to control any equipment, system of at least one operation or part thereof, and
No matter such equipment be hardware, firmware, software or in these at least two certain combination in realize.Should refer to
Go out, the feature being associated with any specific controller can be centralized or distributed, but regardless of being located locally also
It is long-range.The definition for some words and phrase is provided through patent document, and those of ordinary skill in the art will
It is understood that such be defined on many（If not great majority）Be applied in example so definition word and phrase it is first
And the use in future.Although some terms can include various embodiments, claims can understand
These terms are restricted to specific embodiment by ground.
Description of the drawings
In order to the disclosure and its advantage is more fully understood, referring now to the following description being considered in conjunction with the accompanying, wherein phase
Same object is indicated with reference number, and wherein：
Fig. 1 illustrates the block diagram that can wherein realize the data handling system of embodiment；
Fig. 2 illustrates exemplary knee endoprosthesis；
Fig. 3 illustrates the line drawing of the MR images of exemplary knee endoprosthesis and represents；
The flow chart that Fig. 4 illustrates the example process for bone fragments according to the disclosed embodiments；
Fig. 5 illustrates the bone fragments process according to the disclosed embodiments；And
Fig. 6 illustrates twice Iterative classification framework according to the disclosed embodiments.
Specific embodiment
For describe the principle of the disclosure in patent document Fig. 1-6 and each embodiment only as explanation and not
Should be construed to limit the scope of the disclosure by any way.It will be understood by the skilled person that the principle of the disclosure can be with
Realize in the equipment of any appropriate arrangement.Numerous innovative teachings of this application will be described with reference to exemplary non-limiting embodiments.
The disclosed embodiments include for recognize, analyze and classification chart as and element by represented by image it is improved
Imaging system and method, including but not limited to neutralize in other mankind and animal painting for classification and " segmentation " human knee
The imaging system and method for cartilage structure.Although hereafter describing specific embodiment, this area in human knee configuration aspects
It will be recognized that techniques described herein and embodiment go for buttocks, ancon, shoulder, wrist etc.
Deng other anatomical images, and go for other images, described other images include but is not limited to skeleton/cartilage figure
Picture, such as CT scan and other 2D and 3D rendering.
It is useful but challenging task from the automatic segmentation of the human knee cartilage of 3D MR images, this attribution
In the flake structure with disperse border and the cartilage of uneven intensity.
The disclosed embodiments include while be segmented the iteration multiclass learning method of stock cartilage, shin cartilage and kneecap cartilage, and
And can effectively with the spatial context constraint between skeleton and cartilage and between also different cartilages.
Based on the fact cartilage is only grown in some regions of correspondence bone surface, system can be extracted and not only arrive skeleton
The distance feature of the distance on surface, and quantity of information is larger, the distance of the intensive registering anatomic marker in bone surface
Distance feature.
System can also or alternatively using iteration discriminant grader set so that in iteration each time from
The classification confidence figure construction probability comparative feature by derived from the grader for being learnt before.It is emerging that these features are automatically embedded in sense
Semanticss contextual information between the different cartilages of interest.
The disclosed embodiments are included for the fully automated of knee cartilage in 3D MR images, pin-point accuracy and robust
Segmentation method.The method is based on study and effectively with the space constraint between skeleton and cartilage and different
Constraint between cartilage.Specifically, for the prior art in the field, on the surface of correspondence skeleton, institute is intensive
The distance feature of a large amount of anatomic markers of registration, and the classification of the iteration discriminant with probability comparative feature is new and solely
Special.
Although being promoted by the problem of the cartilage segmentation in MRI image, the method for being proposed goes for difference
The general segmentation problem of the different objects in the medical science of mode and other images（Including but not limited to such as GPR, structure,
The field of the x-ray imaging of parcel and container, mm-wave imaging etc.）Effectively to use Semantics of Space information and context
Constrain to lift segmented performance.
Usually, cartilage is not apparent as MRI image.However, the bone learnt by disclosed method
Spatial context constraint between bone and cartilage can be used for jointly constructing the statistical model of skeleton and cartilage.Correspondingly, with
Compared based on more easily estimation can be carried out from the skeleton of CT image segmentations to cartilage with such existing model, which can be with
It is more accurate.After as described herein segmentation or classifying, for the more fully view of motif area, categorical data and
Segmented image can be merged with CT, X-ray or other images.
Fig. 1 illustrates the block diagram that can wherein realize the data handling system of embodiment, such as especially by soft
Part is otherwise configured to perform the imaging system of process as described herein, and especially as retouched herein
Each in the multiple interconnection stated and the system for communicating.The data handling system described includes that being connected to two grades of high speeds delays
/ the processor 102 of bridge 104 is deposited, the second level cache/bridge 104 is subsequently connected to local system bus 106.Local system is total
Line 106 can be, for example, periphery component interconnection（PCI）Framework bus.In illustrated example, local system is also connected to total
Line 106 is main storage 108 and EGA 110.EGA 110 may be coupled to display 111.
Such as LAN（LAN）/ wide area network/wireless（Such as WiFi）Other ancillary equipment of adapter 112 etc also may be used
To be connected to local system bus 106.Local system bus 106 are connected to input/output by expansion bus interface 114（I/O）
Bus 116.I/O buses 116 are connected to keyboard/mouse adapter 118, disk controller 120 and I/O adapters 122.Disk controller
120 may be coupled to storage device 126, and which can be that any appropriate machine can be used or machine-readable storage medium, bag
Include but be not limited to：Non-volatile, hard-coded type media, such as read only memory（ROM）Or EEPROM
（EEPROM）；Magnetic tape strip unit and user's recordable-type media, such as floppy disk, hard disk drive and compact disc read write
（CD-ROM）Or digital versatile disc（DVD）；And optics, electric or magnetic storage device known to other.
In illustrated example, be also connected to I/O buses 116 is audio frequency adapter 124, speaker（It is not shown）Can
Sound is played to be connected to the audio frequency adapter 124.Keyboard/mouse adapter 118 is provided for such as mouse, tracking
The pointing device of ball, tracking pointer etc.（It is not shown）Connection.
I/O adapters 122 can for example be connected to imaging equipment 128, and which can include being configured to perform such as this paper institutes
Any known image system hardware of the process of description, and specifically can include as known to those skilled in the art
MRI, NMRI and MRT are equipped, and other imaging equipments.
Those of ordinary skill in the art will be appreciated that illustrated hardware can be for specific implementation mode in FIG
Change.For example, the hardware in addition to illustrated hardware or illustrated in replacement, can also use other ancillary equipment, such as
Optical disc drive etc..Illustrated example is provided merely for the purpose explained, and illustrated example is not intended to hint
Limit with regard to the framework of the disclosure.
Include the operation using graphical user interface according to the data handling system or imaging system of embodiment of the disclosure
System.The operating system permits multiple display windows are presented simultaneously in graphical user interface, each of which display window to
The different instances of different application or same application provide interface.Can be by user by pointing device manipulating graphics user interface
Cursor.Light target position can change, and/or generate the event for such as clicking on mouse button etc to activate desired sound
Should.
If through suitably modified, in various commercial steerable systems can be adopted, such as positioned at State of Washington thunder
The product Microsoft Windows of the Microsoft of De MengdeTMA version.According to the disclosure as described change or
Create steerable system.
LAN/WAN/ wireless adapters 112 may be coupled to network 130（It is not the part of data handling system 100）, its
Can be the combination of any public or private data processing system network or network as known to those skilled in the art, including
The Internet.Data handling system 100 can be communicated with server system 140 by network 130, and the server system 140 is not yet
It is the part of data handling system 100, and can be for example embodied as detached data handling system 100.
MRI provides the direct and non-invasive visualization of the whole knee endoprosthesis for including soft cartilaginous tissue.In many situations
In, what research worker was interested is that three different cartilages in human knee joint are segmented：Stock cartilage, shin cartilage and kneecap
Cartilage.These cartilages are respectively attached to femur, tibia and patella.Specifically, it is soft with the stock cartilage and shin as one piece construction
Bone is different, and kneecap cartilage is constituted by two detached：Horizontal kneecap cartilage and middle kneecap cartilage.
As knee cartilage is very thin structure and the concrete surface location for being attached to three knee skeletons, thus grind
Study carefully personnel to be preferably segmented knee skeleton first and the priori of skeleton is incorporated in cartilage fragmentation procedure.
Fig. 2 illustrates exemplary knee endoprosthesis, including femur 202, stock cartilage 204, tibia 206, shin cartilage 208, patella
210 and kneecap cartilage 212.Meniscuss and muscle is eliminated from the figure for the sake of clarity.
Accurate and reproducible quantitative cartilage measures required cartilaginous tissue and from the automatic segmentation of MR images is currently
It is difficult or impossible in system, this is because the heterogeneity of cartilage, small size, low contrast in tissue and shape irregularity
Reason.
Fig. 3 illustrates the line drawing of the MR images of exemplary knee endoprosthesis and represents, including femur 302, tibia 206, kneecap
The region of bone 310 and cartilage 314.
The disclosed embodiments include the voxel classification method based on fully automated study for cartilage segmentation.It is disclosed
Technology include being previously segmented into for the corresponding skeleton in knee endoprosthesis, but be not dependent on skeleton-cartilage interface（BCI）It is clear and definite
Classification.
Alternatively, a large amount of anatomic markers of the disclosed embodiments construction from each voxel to bone surface away from
From feature capturing the spatial relationship between cartilage and skeleton.Due to not requiring that BCI is extracted, so whole framework is simplified simultaneously
And error in classification can be avoided to propagate.The disclosed embodiments can construct many logical features and lift forest（multi-pass
feature-boosting forest）And the distance of intensive flag sign is used, it is as described in greater detail below.
Forest used herein can be implemented as random forest, as known to those skilled in the art.As herein
Used, such random forest is integrated classifier, which includes many decision trees and exports as defeated by individual tree institute
The classification of the pattern of the classification for going out.Usually, it is N and makes the variables number in grader be by making the number of training situation
The use of M and construct each tree.Number m represents the number for determining the input variable of decision-making at the node of tree；M should
Much smaller than M.System by selecting n time from all N number of available training situations with putting back to（I.e. using bootstrapping sample）And select
For the training set of the tree.System uses remaining situation to estimate the error of tree by predicting its classification.For each set
Node, system are randomly chosen m variable, and the decision-making at the node is based on this m variable.System is based on this m in training set
Individual variable and calculate optimum division.Each is set " growth " completely and is not trimmed to about（Normal Tree Classifier can constructed such as
When completed）.When submitting to, the general discussion of such random forest can be in en.wikipedia.org/Random_
With Leo Breiman at forest,Random Forests, Machine Learning 45, No. 1 (2001): 5-
Find in 32, be incorporated by reference into hereby both.For example, it is that the combination for setting predictor makes that Breiman describes random forest
Obtain the value that each tree depends on independent sample and the random vector with the same distribution for all trees in forest.
In addition to the connection between cartilage and skeleton, also there is strong spatial relationship among different cartilages, which is in its other party
It is more often ignored in case.For example, stock cartilage is always located in the top of shin cartilage, and two cartilages are being closed in two skeletons
Touched in the area slid past each other during section is mobile each other.
In order to using the constraint, using automatic contextual techniques, the disclosed embodiments realize that iteration discriminant is classified, and make
Obtain and be used for extracting semanticss contextual feature by many class probability figures that grader before is obtained in iteration each time.Tool
The probability of position and random offset can be compared and calculate difference by body ground, system.With existed according to other technologies
Fixed relative position calculates probability statistics and compares, referred to herein as random offset probability difference（RSPD）Feature these
Feature is computationally more efficient, and for the context of different range is more flexible.
Joint classification and recurrence random forest can be used for solving the problems, such as multiple organ segmentations.Recurrence can be used for prediction and
Estimate organ boundary graph.In additive method, based on the value for returning, output organ boundary graph is rich in quantity of information, but still
It is not pin-point accuracy.
However, the disclosed embodiments attempt for very thin layer structured object to be segmented into cartilage so that do not have in regressand value
How many change.It is encoded to for the classification being subjected to supervision alternatively as the range information of spatial prior, this is for Gao Zhun
Exactness and repeatability carry out learning more effectively.In this context, spatial prior is between multiple objects, more specifically cartilage with
Priori spatial relation between skeleton.
Fig. 4 is illustrated for by all one or more data handling systems as described herein（Individually and jointly
It is referred to as " system "）The flow chart of the example process of performed bone fragments.With regard to each step that the figure is briefly described
It is described more fully hereinafter in.
In certain embodiments, in order to by three knee bone fragments in 3D MR images, system can receive tool first
The MR images for having the skeletal structure of manual note train the set of volume, and which is referred to as " ground truth " skeleton or image, because
The classification of skeleton, cartilage and other structures is known（Step 405）.Ground truth image can be converted to grid by system,
And using relevant point drift（CPD）Method performs point set registration to produce correspondence grid（Step 410）.In some cases,
Can be as the A. Myronenko and X. Song's that are incorporated by reference into herebyPoint Set Registration: Coherent Point Drift, IEEE Transactions on Pattern Analysis and Machine
Intelligence, 32(12):2262-2275,2010 uses CPD method described in 12 months.
For example, as described by Myronenko, point set registration is the key component in many Computer Vision Tasks.Point
The target of collection registration is to assign the correspondence between two point sets and recover a pointto-set map to another conversion.Can be with
By for rigidity and non-rigid point set registration the two CPD method solve many factors, including unknown non-rigid spatial alternation,
Big dimension point set, noise and profile.Such method is considered as the alignment of two point sets of Multilayer networks problem, and
Gauss hybrid models are made by maximizing probability（GMM）The centre of moment（Represent the first point set）It is suitable for data（Second point set）.
System can force the GMM centre of moments coherently to move the topological structure to keep point set as packet.In rigid situation, system can
With by using rigidity parameters to the reparameterization of GMM centroid positions forcing relevant constraint, and derive in any dimension
The closed-form solution of the maximization steps of EM algorithm.In non-rigid situation, system can be by adjusting displacement field
（displacement field）And forcing relevant constraint using the calculus of variations derives optimal mapping.
Then system can train and set up the principal component analysiss for all correspondence grids（PCA）Model.These PCA moulds
Type is statistical shape model, and which is captured by average and the change of the object shapes of correspondence grid representation.From training data/volume
Pca model is set up in set, and and then can by pca model be applied to test data/volume for segmentation purpose.
Then system can combine pca model to produce the active shape model including mean shape and change of shape space
（ASM）, average shape of the mean shape corresponding to the skeleton and cartilage of training volume, change of shape space is corresponding to training volume
Skeleton and cartilage orientation change change of shape space.Then system can make these ASM be suitable for image to determine
The initial fragment of three knee skeletons（Step 420）Although coordinating the segmentation for carrying out usually inaccurate by the initial ASM
's.
In the T. Coots being incorporated by reference into hereby, C. Taylor, D. Cooper and J. Graham'sActive Shape Models – Their Training and Application, Computer Vision and Image
Understanding, 61(1):ASM technologies are described in 38-59,1995.For example, Coots describes a kind of for passing through
The method that study forms the variable sexual norm of the training set of the correct image explained and sets up model.These ASM can be used for
Image in iterative refinement algorithm searches element, while only in the way of causing with training set unification deforming ASM.
Then system can morphologically corrode mask with the skeletogenous positive seed of life, make mask expansion skeletogenous with life
Negative seed, and perform random walk process to obtain the bone fragments mask of refinement（Step 425）.Hereby by quoting simultaneously
The L. Grady for entering,Random Walks for Image Segmentation, IEEE Transactions on
Pattern Analysis and Machine Intelligence, 28(11):Described in 1768-1783, Nov. 2006
A kind of suitable random walk algorithm.For example, Grady describes a kind of process, gives and limits with user（Or limit in advance
It is fixed）Labelling a small amount of pixel, the process analytically and can quickly determine and open at each unlabelled pixel
The random walk of beginning will arrive first at the probability of one of the pixel of advance labelling.By to the labelling for its calculating maximum of probability
Assign each pixel, it is possible to obtain high-quality image segmentation.Such process can be used from continuous potential theoretical mark
The combine analog of quasi- operator and principle and in discrete space（For example on figure）Perform, so as to allow to be applied on Subgraph
Any dimension.
It is signed to construct that then system can perform the signed range conversions of 3D on refinement bone fragments mask
To skeleton distance feature（Step 430）.Then system can apply CPD processes again to produce the corresponding grid of segmentation result
（Step 435）.System is configured to intensive tag distance feature using these correspondence grids（Step 440）.
As mentioned above, first for two main causes by knee bone fragments.First, bone surface is soft
Bone segmentation provides important space constraint.Secondly, comparatively it is easier bone fragments, because they have more regular and distinguish
The shape of property.
The disclosed embodiments pass through closed triangle gridThe shape of skeleton is represented, whereinBe N number of mesh point set andIt is the set of M triangle index.It is given
With multiple training volumes that manual skeleton is explained, system can find the anatomy correspondence of mesh point simultaneously using CPD processes
And which constructs to have and is expressed asMean shape statistical shape model.
Fig. 5 illustrates the bone fragments process according to the disclosed embodiments, and which includes training stage 500 and detection-phase
520.Training stage 500 is included at 502 and receives the training volume with the skeleton and cartilage explained manually（Ground truth figure
Picture）, CPD processes are performed at 503 to produce correspondence grid at 504, and perform PCA processes to produce at 506 at 505
Raw active shape model, the active shape model include mean shape and corresponding change of shape space.
Detection-phase 500 is using the active shape model produced at 506.These are applied to each MR to be processed
Image（" model " or " test model "）, they are received by system, to determine test volume at 522.Test volume is will quilt
It is categorized as the region of the test model of skeleton, cartilage or other forms." region " or " portion of MR test models described herein
Point " it is intended to refer to 2D regions or 3D volumes.
System at 523 uses Attitude estimation by marginal space learning to produce estimated by model at 524
Translation, rotation and scaling so that mean shape and change of shape space of the test volume corresponding to active shape model.
System by 525 using the translation estimated by model, rotation and scaling, via according to mean shape and shape
The iterative boundary cooperation that shape change space makes test volume be suitable for active shape model and carries out carrys out application model deformation.This
Initial fragment is produced at 526.
Initial fragment is the part of test model to skeleton, cartilage or other preliminary classifications.
Then system performs random walk process using initial fragment on the model for border refinement at 527, with
Refinement segmentation is produced at 528.In this process, system refines the border of test volume with regard to active shape model.Refinement
Segmentation is the part of test model to skeleton, cartilage and other classification.Then system can refine fragmented storage
The categorical data for answering test model associated, the classification or segmentation of each part of the categorical data mark test model.Should
Process can also include showing MR image measurement models, including the MR image measurement moulds of the classification indicated corresponding to each voxel
The part of type.As it is used herein, can include from storage device loading or from another equipment or process by system " reception "
Receive.
In this embodiment, whole bone fragments framework can include three key step " Attitude estimations（To determine model
Estimated translation, rotation and scaling）, model deformation（With the translation of rightly application model, rotation and scaling）And border
Refinement（Test volume is made to be adapted to active shape model to adjust border）.
Attitude estimation：For given volume V, first by searching plain optimum posture parameterTo position skeleton, whereinRepresent translation,Rotation is represented, andRepresent anisotropy scaling.
In order to accelerate detection, adopt and be referred to as marginal space learning（MSL）Efficient derivation with will be initial parameter empty
Between in exhaustive search resolve into three decision estimation problems, each in compared with low dimensional limit space in, such as by following table
Show：
Mean shape is converted linearly by using estimated attitude parameterTo initialize shape.
In the Y. Zheng being incorporated by reference into hereby, A. Barbu, M. Georgescu, M. Scheuring and
D. Four-Chamber Heart Modeling and Automatic Segmentation for 3D of Comaniciu
Cardiac CT Volumes Using Marginal Space Learning and Steerable Feature. IEEE
Trans. Med. Imag., 27(11):Acceptable MSL processes are described in 1668-1681,2008.For example, ZHeng is retouched
The MSL processes for solving 9-D similarity transformation's search problems of positioning ventricle are stated.It is determined that after the attitude of ventricle, being somebody's turn to do
Process estimates 3-D shapes by based on the border delineation of study.The MSL processes can be by will estimate to be divided into three problems
And incrementally learn to project the grader on sample distribution：The conversion of location estimation, position orientation estimation and absolute similarity is estimated.
Such method can also use based on the 3-D edge detectors of study to guide the shape distortion in ASM frameworks.
Model deformation：In the stage, by along normal direction search and mobile each mesh pointExtremely
New point with the maximum probability generated by the set of border grader make original shape deform to be suitable for border.It is all
The structural strain's of mesh point are projected to the change subspace of set up statistical shape model.If the boundary fitting process repeats
Dry time till convergence.
Border refines：In order to further improve segmentation accuracy, system refines bone boundaries using random walk process.
The anatomy correspondence of mesh point may be lost in the stage.Thus, system using CPD processes come before being registered in random walk and
Two grid point sets afterwards with obtain refinement bone surface on the mark being anatomically equal to, its will be used for extract away from
From feature, as will be described in more detail.
System can perform cartilage classification, and which is related to feature extraction, iteration semanticss contextual feature and is lifted, and passed through
The post processing of figure cutting optimization.Given all three knee skeleton is segmented, and system is extracted most from each bone surface first
Band interested in big distance threshold.This can linear session by the range conversion to binary skeleton mask efficiently
Realize.By only classifying to the voxel in band interested, not only significantly reduce the calculating cost for test, but also
Training is simplified by removing incoherent negative voxel.
Feature extraction：For each voxel with space coordinatess x, system can construct multiple basic features, which can
To classify as three subsets.Strength characteristic includes voxel intensities and its gradient magnitude respectively：.Distance feature measures each voxel to the tape symbol Europe of different knee bone boundaries
Distance is obtained in several：, whereinIt is the tape symbol to femur
Distance,Be to tibia, andIt is to patella.System also uses linear combination：
These be characterized in that it is useful because summation feature f6And f8Whether tolerance voxel x is located at the narrow sky between two skeletons
In, and difference feature f7And f9Which is measured closer to which skeleton.Except strength characteristic f1Outside, feature f of suing for peace6And f7Shin is soft
Bone is separated with stock cartilage and kneecap cartilage.
The priori grown in some regions that given cartilage only can be in bone surface, which is for cartilage segmentation
Speech usefully not only knows how voxel is close to bone surface, and knows which anatomically in where.Therefore, system
Using as described herein to the following distance feature of the intensive registration marks in bone surface：
WhereinIt is all skeleton mesh pointsThe space coordinatess of mark.It is attributed to a large amount of mesh points can use,It is
The random generation in the training of grader.
The intensity of current voxel x and another voxel x+u is compared by contextual feature with random deviation u：
Wherein u is random deviation vector.Referred to herein as " random offset strength difference "（RSID）The spy of feature
Levy subset the contextual information in different range is captured by generating a large amount of different value u in training at random.These features can
Problem is recognized for solving attitude classification and key point.
Iteration semanticss contextual feature is lifted：The disclosed embodiments can be using multipass Iterative classification process come automatically
Ground is with the semanticss context for multiple object fragments problems.Each in, the probability graph for being generated will be used for extracting
The embedded feature of context is lifting the classification performance of next time.
Fig. 6 illustrates two times Iterative classification mistakes of the random forest with the basic classification device for being selected for use in each time
Journey, but the method can be extended to using the more all over iteration of other discriminant graders.
The cartilage fragmentation procedure includes two-level classifier, is also known as twice classification.The embodiment includes the training stage 600
With test phase 630, they can perform simultaneously or independently can perform.That is, having a period, training stage
600 can complete and its result is stored, and in the subsequent time, it is possible to use the result of training stage 600 performs survey
The examination stage 630.
Set of the training stage 600 based on training image 602 and the cartilage ground explained in each training image manually
The comparison of face live state information 604.Training output be the first pass with selected feature, threshold value, class probability and other specification with
Machine forest classified device 610.First pass random forest grader 610 can be used for producing first pass probability graph 606.
Training stage 600 and then second time can be included, wherein cartilage ground truth information 604 and first pass probability graph
606 compare to produce second time random forest grader 620 with selected feature, threshold value, class probability and other specification.
Second time random forest 620 can be used for producing second time probability graph 608.
During test phase 630, for any test image 632（Without the MR image measurement moulds of ground truth
Any given image of type）, these random forest graders can be at 633 for automatically by each voxel in image
One be categorized as in four classification：Stock cartilage, shin cartilage, kneecap cartilage and background（Other anything）.
More specifically, each test image at 633 compared with first pass random forest grader 610 producing one
As there is the first pass probability graph 634 of same size with input picture.The value of each voxel in first pass probability graph 634
Indicate the voxel as the probability of any one in four classification.Usually, voxel will have in being classified as four classification
One of maximum likelihood.
The first pass probability graph 634 obtained by first pass random forest grader 610 can be by second time random forest
Grader 620 at 635 is used for improving classification performance.The second of second time acquisition of random forest grader 620 is used at 635
Final classification decision-making be can serve as all over probability graph 636, and system then can be corresponding to the classification number of second time probability graph
It is stored as what is be associated with MR image measurement models according to 640.
Part finally, as test phase 630 or discretely, system can at 650 to test image in it is every
One is segmented.The process can include using categorical data 640 for each test image being segmented into zones of different, all
Bony areas, cartilaginous areas and other or background area as, or concrete region, such as patellar region, kneecap cartilage area
Domain, femoral region, stock cartilaginous areas, tibial region and shin cartilaginous areas.The process can include depositing together with categorical data
The segment data or storage are stored up as the segment data of the part of categorical data, or can include noting using segment data
Each test image is solved using as metadata or other forms so that when image is displayed to user, corresponding segments
According to can also be shown with labelling or be otherwise indicated that（Such as pass through color coding）Different sectional areas.Test image is arrived
Each voxel can be assigned to respective regions based on the classification according to the voxel by the segmentation of zones of different.
Certainly, system can be extended to the grader more than two-stage, and each of which level is general using what is obtained from previous stage
Rate figure, and the probability graph from afterbody output is used as final classification decision-making.
Semanticss contextual feature：Each after classification, generating probability figure and probability graph are used for extracting semanticss
Contextual feature, limits as following：
WhereinWithStock cartilage, shin cartilage and kneecap cartilage probability graph is represented respectively.With with RSID features above
The probability respondence of two voxels and random offset can be compared by identical mode, system
Which is referred to as random offset probability difference characteristic（RSPD）, there is provided semanticss contextual information, because strong with original
Degree volume compares, and probability map values are directly associated with anatomic landmarks.
In such multipass sort system, the probability of each subsequent passes is illustrated with the quantitative of less noise response
Improve.
By scheming the post processing that cutting optimization is carried out：After multipass Iterative classification, system can be classified using four（The back of the body
Scape, stock cartilage, shin cartilage and kneecap cartilage）Probability, with construct energy function and perform multi-signature cutting it is smooth to utilize
Spend constraint to refine segmentation result.In the Y. Boykov being incorporated by reference into hereby, O. Veksler and R. Zabih.'sFast Approximate Energy Minimization via Graph Cuts, IEEE Transactions on
Pattern Analysis and Machine Intelligence, 23(11):Described in 1222-1239, Nov. 2001
Acceptable figure cutting process.For example, Boykov describes mobile and exchange movement with regard to extension based on efficiently finding
The process of the figure cutting of local minimum.These movements may change the labelling of any big collection pixel simultaneously.Such process
Find the labelling in the known facts of global minimum and dispose general energy function.These processes allow discontinuous holding energy
The important condition of amount.
Labelling l (x) is assigned to each voxel x and labelling l (y) is assigned to voxel y by figure cutting algorithm so that with
Under ENERGY E minimize：
Wherein L is global mark configuration, and N is adjacent system,It is smoothness energy, andIt is data energy
Amount.System specialization
。
The value 1 when l (x) and l (y) are not isolabeling, and the value 0 in l (x)=l (y).ValueOr, this depends on labelling l (x).With
It is two parameters.Specified data capacity contrasts the weight of smoothness energy, andRepresent picture noise.
Certainly, it would be recognized by those skilled in the art that unless the sequence of operation is specifically indicated or required, otherwise being retouched above
Some steps during stating can be omitted, concurrently or in turn perform or performed with different order.It is described herein
The various features and process of embodiment can be in combination with one another in the scope of the present disclosure.
It would be recognized by those skilled in the art that for simplicity and clearness, being suitable to all data used for the disclosure
The entire infrastructure of processing system and operation are not illustrated herein.Alternatively, only describe and describe it is unique for the disclosure or
Person is for the so much content for understanding the necessary data handling system of the disclosure.Data handling system 100 is construction and operation of
Remainder can meet any one in various Current implementations as known in the art and practice.
Although it is important to note that the disclosure includes the description in the context of complete function system, this area skill
Art personnel by, it is realized that the disclosure at least partly be able to being used by being included in any various forms of machines of mechanism, in terms of
Calculation machine can use or computer-readable medium in instruction formal distribution, and the disclosure is equally suitable for but regardless of being used for real
Carry out the particular type of the instruction of the distribution or signal bearing medium or storage medium in border.Machine can be using/readable or computer
Can be included using the example of/computer-readable recording medium：Non-volatile, hard-coded type media, such as read only memory（ROM）Or electrically erasable can
Program read-only memory（EEPROM）；And user's recordable-type media, such as floppy disk, hard disk drive and compact disk is read-only deposits
Reservoir（CD-ROM）Or digital versatile disc（DVD）.
Although have been described in the exemplary embodiment of the disclosure, it will be understood by those skilled in the art that can
To make various changes, replacement, modification and improvement without departing from the disclosure in the case of the spirit and scope of its most wide form.
In the application, no any description should be read as implying that any particular element, step or function are to include
Requisite item in right：It is required that the scope of the theme of patent right is only limited by the claim for authorizing.
Claims (20)
1. a kind of method for being classified to the skeleton and cartilage in magnetic resonance image (MRI), including：
- in a data processing system receive magnetic resonance image (MRI) test model, the test model include non-classified skeleton and
Cartilage portion；
- test volume of the test model is determined by the data handling system, the test volume includes the test
The skeleton to be categorized as of model or the region of cartilage；
- test model is changed by the data handling system so that the test volume corresponds to active shape model
Mean shape and change of shape space；
- by the data handling system, it is suitable for the average of the active shape model by making the test volume
Shape and the change of shape space and produce the test volume to bony areas and the preliminary classification of cartilaginous areas；
- by the data handling system, produced by refining the border of the test volume with regard to the active shape model
The life test volume is to bony areas and the classification of cartilaginous areas；And
- the magnetic resonance image (MRI) test model is segmented into corresponding to bony areas and cartilaginous areas according to the classification
Zones of different.
2. the method for claim 1, also includes：
- multiple training volumes are received, the plurality of training volume includes the known region of skeleton and cartilage；
- perform coherent point Drift Process to produce correspondence grid, the correspondence grid each corresponding to it is corresponding each
Training volume；
- perform principal component analysiss process to produce principal component model from the correspondence grid；And
- active shape model is produced from the principal component model.
3. method as claimed in claim 2, wherein described coherent point Drift Process are included in the plurality of training volume and hold
Row point set registration.
4. the method for claim 1, wherein changing the test model includes for model deformation being applied to the test
Model, the model deformation include translation, rotation and the scaling of the test model.
5. the method for claim 1, wherein described data handling system are estimated using attitude by marginal space learning
The translation of the estimation required by modification for counting to produce the test model, rotation and scaling.
6. the method for claim 1, wherein refines the border bag of the test volume with regard to the active shape model
Include execution random walk process.
7. the method for claim 1, wherein described magnetic resonance image (MRI) test model is the magnetic resonance image (MRI) of human knee.
8. the method for claim 1, is wherein segmented into the magnetic resonance image (MRI) test model corresponding to bony areas
Include for the magnetic resonance image (MRI) test model being segmented into patellar region, kneecap cartilaginous areas, stock with the zones of different of cartilaginous areas
Bone region, stock cartilaginous areas, tibial region and shin cartilaginous areas.
9. the method for claim 1, wherein described data handling system is using the segmentation corresponding to the zones of different
Data are explaining the magnetic resonance test image model.
10. the method for claim 1, wherein described data handling system show the magnetic resonance image (MRI) test model,
Including the zones of different indicated on the magnetic resonance image (MRI) test model.
A kind of 11. data handling systems, including：
Processor；And
Addressable memorizer, the data handling system are configured to：
- magnetic resonance image (MRI) test model is received, the test model includes non-classified skeleton and cartilage portion；
- determine the test volume of the test model, the test volume include the skeleton to be categorized as of the test model or
The region of cartilage；
- modification the test model causes the test volume corresponding to the mean shape and change of shape sky of active shape model
Between；
- test volume is produced to bony areas and the preliminary classification of cartilaginous areas；
- test volume is produced to skeleton area by refining the border of the test volume with regard to the active shape model
Domain and the classification of cartilaginous areas；And
- the magnetic resonance image (MRI) test model is segmented into corresponding to bony areas and cartilaginous areas according to the classification
Zones of different.
12. data handling systems as claimed in claim 11, wherein described data handling system are further configured to：
- multiple training volumes are received, the plurality of training volume includes the known region of skeleton and cartilage；
- perform coherent point Drift Process to produce correspondence grid, the correspondence grid each corresponding to it is corresponding each
Training volume；
- perform principal component analysiss process to produce principal component model from the correspondence grid；And
- active shape model is produced from the principal component model.
13. data handling systems as claimed in claim 12, wherein described coherent point Drift Process are included in the plurality of instruction
Practice.
14. data handling system as claimed in claim 11, wherein changing the test model is included model deformation application
In the test model, the model deformation includes translation, rotation and the scaling of the test model.
15. data handling systems as claimed in claim 11, wherein described data handling system is by marginal space learning
Translation, rotation and the scaling of estimation required by the modification of the test model is produced using Attitude estimation.
16. data handling systems as claimed in claim 11, wherein refine the test body with regard to the active shape model
Long-pending border includes performing random walk process.
17. data handling systems as claimed in claim 11, wherein described magnetic resonance image (MRI) test model is human knee
Magnetic resonance image (MRI).
The magnetic resonance image (MRI) test model is wherein segmented into correspondence by 18. data handling systems as claimed in claim 11
Zones of different in bony areas and cartilaginous areas includes for the magnetic resonance image (MRI) test model being segmented into patellar region, kneecap soft
Bone region, femoral region, stock cartilaginous areas, tibial region and shin cartilaginous areas.
19. data handling systems as claimed in claim 11, wherein described data handling system is using corresponding to the difference
The segment data in region is explaining the magnetic resonance test image model.
20. data handling systems as claimed in claim 11, wherein described data handling system show the magnetic resonance image (MRI)
Test model, including the zones of different indicated on the magnetic resonance image (MRI) test model.
Applications Claiming Priority (5)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US201261734280P | 2012-12-06 | 2012-12-06 | |
| US61/734280 | 2012-12-06 | ||
| US13/776184 | 2013-02-25 | ||
| US13/776,184 US9218524B2 (en) | 2012-12-06 | 2013-02-25 | Automatic spatial context based multi-object segmentation in 3D images |
| PCT/US2013/073614 WO2014089455A2 (en) | 2012-12-06 | 2013-12-06 | Automatic spatial context based multi-object segmentation in 3d images |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104956397A CN104956397A (en) | 2015-09-30 |
| CN104956397B true CN104956397B (en) | 2017-03-22 |
Family
ID=50881019
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201380072420.4A Active CN104956397B (en) | 2012-12-06 | 2013-12-06 | Automatic spatial context based multi-object segmentation in 3D images |
Country Status (5)
| Country | Link |
|---|---|
| US (1) | US9218524B2 (en) |
| EP (1) | EP2929509B1 (en) |
| JP (1) | JP6005297B2 (en) |
| CN (1) | CN104956397B (en) |
| WO (1) | WO2014089455A2 (en) |
Families Citing this family (32)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014145691A1 (en) | 2013-03-15 | 2014-09-18 | Otismed Corporation | Generation of a mating surface model for patient specific cutting guide based on anatomical model segmentation |
| US9495752B2 (en) * | 2012-09-27 | 2016-11-15 | Siemens Product Lifecycle Management Software Inc. | Multi-bone segmentation for 3D computed tomography |
| EP3327601B1 (en) | 2014-01-27 | 2020-12-09 | Materialise N.V. | Prediction of shapes |
| US9367924B2 (en) * | 2014-05-06 | 2016-06-14 | Siemens Aktiengesellschaft | Method and system for segmentation of the liver in magnetic resonance images using multi-channel features |
| US9710880B2 (en) * | 2014-07-03 | 2017-07-18 | Siemens Product Lifecycle Management Software Inc. | User-guided shape morphing in bone segmentation for medical imaging |
| WO2016007518A1 (en) * | 2014-07-07 | 2016-01-14 | The Regents Of The University Of California | Automatic segmentation and quantitative parameterization of brain tumors in mri |
| JP6482250B2 (en) * | 2014-11-20 | 2019-03-13 | キヤノン株式会社 | Image processing apparatus, image processing method, and program |
| KR102361733B1 (en) * | 2014-11-28 | 2022-02-11 | 삼성전자주식회사 | Method and apparatus for modeling structure of coronary artery from 3d coronary ct angiography(cta) image |
| US10037603B2 (en) * | 2015-05-04 | 2018-07-31 | Siemens Healthcare Gmbh | Method and system for whole body bone removal and vascular visualization in medical image data |
| CN107709930B (en) * | 2015-06-18 | 2021-08-31 | 宝马股份公司 | Method and device for representing map elements and method and device for locating vehicle/robot |
| CN108351396B (en) * | 2015-11-02 | 2021-09-14 | 皇家飞利浦有限公司 | Method, computer program product and magnetic resonance imaging system for tissue classification |
| US9801601B2 (en) | 2015-12-29 | 2017-10-31 | Laboratoires Bodycad Inc. | Method and system for performing multi-bone segmentation in imaging data |
| US9898858B2 (en) * | 2016-05-18 | 2018-02-20 | Siemens Healthcare Gmbh | Human body representation with non-rigid parts in an imaging system |
| JP6687894B2 (en) * | 2016-05-20 | 2020-04-28 | 富士ゼロックス株式会社 | Class estimation device and program |
| JP6933521B2 (en) * | 2016-08-05 | 2021-09-08 | キヤノンメディカルシステムズ株式会社 | Medical image processing equipment and X-ray CT equipment |
| US10726608B2 (en) | 2016-11-23 | 2020-07-28 | 3D Systems, Inc. | System and method for real-time rendering of complex data |
| CN108205805B (en) * | 2016-12-20 | 2020-06-02 | 北京大学 | Automatic establishment method for dense correspondence of cone beam CT image intermediate pixels |
| JP6685250B2 (en) * | 2017-02-24 | 2020-04-22 | 富士フイルム株式会社 | Mapping image generation device, method and program |
| US10706554B2 (en) | 2017-04-14 | 2020-07-07 | Adobe Inc. | Three-dimensional segmentation of digital models utilizing soft classification geometric tuning |
| US10667867B2 (en) | 2017-05-03 | 2020-06-02 | Stryker European Holdings I, Llc | Methods of pose estimation of three-dimensional bone models in surgical planning a total ankle replacement |
| US10813614B2 (en) * | 2017-05-24 | 2020-10-27 | Perkinelmer Health Sciences, Inc. | Systems and methods for automated analysis of heterotopic ossification in 3D images |
| US11903691B2 (en) * | 2017-07-18 | 2024-02-20 | Siemens Healthineers Ag | Combined steering engine and landmarking engine for elbow auto align |
| EP3493154A1 (en) * | 2017-12-01 | 2019-06-05 | Koninklijke Philips N.V. | Segmentation system for segmenting an object in an image |
| US11302008B2 (en) * | 2018-03-30 | 2022-04-12 | Perkinelmer Health Sciences, Inc. | Systems and methods for automated detection and segmentation of vertebral centrum(s) in 3D images |
| CN108805203A (en) * | 2018-06-11 | 2018-11-13 | 腾讯科技（深圳）有限公司 | Image procossing and object recognition methods, device, equipment and storage medium again |
| US11257214B2 (en) | 2018-12-03 | 2022-02-22 | Regents Of The University Of Minnesota | Two-stage variational image segmentation of medical images using fracture mechanics |
| CN110827294A (en) * | 2019-10-31 | 2020-02-21 | 北京推想科技有限公司 | Network model training method and device and focus area determination method and device |
| EP3828768A1 (en) | 2019-11-28 | 2021-06-02 | Digital Orthopaedics | Method for modeling a bone |
| EP3866107A1 (en) * | 2020-02-14 | 2021-08-18 | Koninklijke Philips N.V. | Model-based image segmentation |
| WO2023026115A1 (en) * | 2021-08-25 | 2023-03-02 | Medx Spa | Automated quantitative joint and tissue analysis and diagnosis |
| CN115861571B (en) * | 2023-01-18 | 2023-04-28 | 武汉大学 | Semantic perception triangle network model building entity reconstruction method |
| CN116452755B (en) * | 2023-06-15 | 2023-09-22 | 成就医学科技(天津)有限公司 | Skeleton model construction method, system, medium and equipment |
Family Cites Families (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8556983B2 (en) * | 2001-05-25 | 2013-10-15 | Conformis, Inc. | Patient-adapted and improved orthopedic implants, designs and related tools |
| US7239908B1 (en) * | 1998-09-14 | 2007-07-03 | The Board Of Trustees Of The Leland Stanford Junior University | Assessing the condition of a joint and devising treatment |
| US7764838B2 (en) * | 2004-09-14 | 2010-07-27 | Siemens Medical Solutions Usa, Inc. | System and method for extracting an object of interest from an image using a robust active shape model |
| CN101150983B (en) * | 2005-01-28 | 2015-01-07 | 加州大学董事会 | Systems and methods using nuclear magnetic resonance (NMR) spectroscopy to evaluate pain and degenerative properties of tissue |
| US7831078B2 (en) * | 2006-07-24 | 2010-11-09 | Siemens Medical Solutions Usa, Inc. | System and method for statistical shape model based segmentation of intravascular ultrasound and optical coherence tomography images |
| EP2067121A2 (en) * | 2006-09-19 | 2009-06-10 | Nordic Bioscience A/S | Pathology indicating measure related to cartilage structure and automatic quantification thereof |
| US8116548B2 (en) * | 2008-06-04 | 2012-02-14 | Siemens Aktiengesellschaft | Method and system for detecting 3D anatomical structures using constrained marginal space learning |
| EP2385791B1 (en) * | 2008-08-12 | 2014-03-12 | Wyeth Pharmaceuticals Inc. | Morphometry of the human hip joint and prediction of osteoarthritis |
| US8411986B2 (en) * | 2009-04-13 | 2013-04-02 | Flashfoto, Inc. | Systems and methods for segmenation by removal of monochromatic background with limitied intensity variations |
| US8306257B2 (en) * | 2011-01-31 | 2012-11-06 | Seiko Epson Corporation | Hierarchical tree AAM |
| US8422754B2 (en) * | 2011-02-15 | 2013-04-16 | Siemens Aktiengesellschaft | Medical imaging system |
| KR101842043B1 (en) * | 2011-07-22 | 2018-03-26 | 삼성전자주식회사 | Apparatus and Method for analyzing ultrasound images |
| CA2845044C (en) * | 2011-08-12 | 2023-03-28 | Jointvue, Llc | 3-d ultrasound imaging device and methods |
| US10388020B2 (en) * | 2012-04-11 | 2019-08-20 | The Trustees Of Columbia University In The City Of New York | Methods and systems for segmentation of organs and tumors and objects |
| US9747527B2 (en) * | 2013-03-15 | 2017-08-29 | Nvidia Corporation | Performing object detection operations via random forest classifier |
-
2013
- 2013-02-25 US US13/776,184 patent/US9218524B2/en active Active
- 2013-12-06 CN CN201380072420.4A patent/CN104956397B/en active Active
- 2013-12-06 EP EP13811103.4A patent/EP2929509B1/en active Active
- 2013-12-06 WO PCT/US2013/073614 patent/WO2014089455A2/en active Application Filing
- 2013-12-06 JP JP2015545879A patent/JP6005297B2/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN104956397A (en) | 2015-09-30 |
| JP2016505298A (en) | 2016-02-25 |
| JP6005297B2 (en) | 2016-10-12 |
| EP2929509B1 (en) | 2017-06-14 |
| US9218524B2 (en) | 2015-12-22 |
| WO2014089455A2 (en) | 2014-06-12 |
| EP2929509A2 (en) | 2015-10-14 |
| WO2014089455A3 (en) | 2014-08-21 |
| US20140161334A1 (en) | 2014-06-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104956397B (en) | Automatic spatial context based multi-object segmentation in 3D images | |
| Ramesh et al. | A review of medical image segmentation algorithms | |
| Mharib et al. | Survey on liver CT image segmentation methods | |
| Poudel et al. | Evaluation of commonly used algorithms for thyroid ultrasound images segmentation and improvement using machine learning approaches | |
| Seim et al. | Automatic Segmentation of the Pelvic Bones from CT Data Based on a Statistical Shape Model. | |
| CN111640120B (en) | Pancreas CT automatic segmentation method based on significance dense connection expansion convolution network | |
| Cai et al. | Multi-modality vertebra recognition in arbitrary views using 3D deformable hierarchical model | |
| Chen et al. | GC-ASM: Synergistic integration of graph-cut and active shape model strategies for medical image segmentation | |
| Wu et al. | Segmentation of multiple knee bones from CT for orthopedic knee surgery planning | |
| Tripathi et al. | Image segmentation: A review | |
| Wang et al. | Semantic context forests for learning-based knee cartilage segmentation in 3D MR images | |
| Zografos et al. | Hierarchical multi-organ segmentation without registration in 3D abdominal CT images | |
| Wu et al. | Lung segmentation based on customized active shape model from digital radiography chest images | |
| Gsaxner et al. | PET-train: Automatic ground truth generation from PET acquisitions for urinary bladder segmentation in CT images using deep learning | |
| Tong et al. | Segmentation of brain MR images via sparse patch representation | |
| CN114882051A (en) | Automatic segmentation and three-dimensional reconstruction method for pelvic bone tumor based on multi-modal image | |
| Donner12 et al. | Sparse MRF appearance models for fast anatomical structure localisation | |
| Ridhma et al. | Review of automated segmentation approaches for knee images | |
| Oktay et al. | Probabilistic edge map (PEM) for 3D ultrasound image registration and multi-atlas left ventricle segmentation | |
| Birkbeck et al. | Robust segmentation of challenging lungs in CT using multi-stage learning and level set optimization | |
| Platero et al. | A new label fusion method using graph cuts: application to hippocampus segmentation | |
| Li et al. | An overview of abdominal multi-organ segmentation | |
| Dan et al. | Multi-scale adaptive level set segmentation method based on saliency | |
| Park et al. | Automatic bone segmentation in knee MR images using a coarse-to-fine strategy | |
| Fischer et al. | Structural scene analysis and content-based image retrieval applied to bone age assessment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CP01 | Change in the name or title of a patent holder | ||
| CP01 | Change in the name or title of a patent holder |
Address after: American Texas Patentee after: SIEMENS INDUSTRY SOFTWARE N.V. Address before: American Texas Patentee before: SIEMENS PRODUCT LIFECYCLE MANAGEMENT SOFTWARE Inc. |